Comparison
oruk vs Hume AI
Compare speech-understanding contracts before moving your audio pipeline. Oruk analyzes speech; an EVI migration also needs a language model, speech generation and conversation orchestration.
API plans from $9/month. Choose a model by language, output and timing requirements.
Scores describe vocal expression, not private feelings. Oruk and Hume use different label sets; compare them on your own audio.
Product information reviewed . Benchmark snapshots retain their original dates.
Hume EVI and TTS have an announced cutoff
Hume says EVI and TTS API access ends November 13, 2026 at 12:01 a.m. EST (November 12, 2026 at 9:01 p.m. PST). Complete and verify your replacement before that time.
Hume’s notice says account data is permanently deleted after November 13, 2026. Export the data you need and verify your archive before access ends.
Read Hume’s official noticeoruk is not affiliated with, endorsed by, or sponsored by Hume AI, Inc. “Hume” and product names are trademarks of their owner, used here for identification and comparison only. Claims about Hume products are sourced below and dated; if anything is out of date, email access@oruk.ai and we will correct it.
Correction — July 20, 2026: an earlier version treated a legacy Hume Expression Measurement snapshot as evidence that Hume had left the category. Hume’s current official product page describes offline Tagger and real-time Prosody products. The comparison below now separates those products from the legacy API snapshot in our benchmark.
Hume’s Expression Measurement page separately advertises offline Tagger and real-time Prosody, with contact-led access. The EVI/TTS sunset notice does not establish that these products are ending. Confirm their availability and terms with Hume Research. Oruk’s models have separate file and WebSocket contracts; choose one using the migration compatibility guide. The 49.6% Hume benchmark result below belongs to a legacy prosody snapshot, not current Tagger or Prosody.
Side by side
| oruk | Hume AI | |
|---|---|---|
| Product scope (reviewed October 3, 2026) | Speech understanding: English timed file analysis, short multilingual recordings, acoustic scores and live phrase events through separate contracts | Tagger/Prosody measurement; a separate announced cutoff for EVI and TTS (see notice above) |
| Offline emotion measurement | Resonance 1 for English timed analysis; Spectra-2 for short multilingual clips; Resonance 2 for expression scores without a transcript | Tagger — batch analysis; Hume currently directs users to contact Research for access |
| Emotion output | 15 emotion and 16 style categories; selected labels, full scores and timing depend on the model. Realtime phrase labels are variable-length | Current Tagger advertises 600+ expression dimensions; the legacy prosody endpoint returned 48 |
| speech-emotion-bench accuracy | 77.8% for the July 2026 Resonance 1 checkpoint (trained in-distribution; not a new evaluation of later builds) | 49.6% for the legacy 48-dimension prosody snapshot on the 5,000-clip subset; current Tagger/Prosody not tested |
| Input | Resonance 1: English files. Spectra-2: 25 transcription languages. Realtime: 32 transcription locales. Resonance 2: acoustic clips; validate performance by language | Recorded audio through Tagger; live audio through Prosody/EVI; text for TTS |
| Streaming / real time | Realtime: transcript deltas + phrase events. Resonance 2: explicit range commits. Spectra-2: upload then finish, with a final result | Prosody is advertised for live measurement; EVI combines a voice-agent stack and is covered by the sunset notice |
| Pricing model | Plans cover the speech model lineup. New organizations need separate approval for Resonance 2; an active trial or plan does not by itself grant access. Hobby is $9/month, Builder $49/month, and Production $199/month. Spectra-2, Resonance 1, Resonance 2, Fourier, Realtime, and Proficiency share 250, 2,500, or 20,000 monthly audio minutes, respectively, including supported diarization. Each plan also includes a separate Orukeet allowance, starting at up to 20,000 transcription minutes on Hobby. Orukeet uses $0.00045/minute ($0.027/hour); optional tasks draw from its allowance at their published rates. | Current Tagger/Prosody access and pricing: contact Hume Research |
Benchmark numbers are same-harness measurements from speech-emotion-bench (77.8% oruk Resonance 1 vs 49.6% Hume prosody); the oruk result describes the dated July 2026 checkpoint, not a new evaluation of subsequent API builds or Fourier. The oruk entry is trained in-distribution and the Hume legacy API row was scored on the fixed 5,000-clip subset. Open models use the full 64,384-clip evaluation. The caveats and protocol are on the methodology page.Plans cover the speech model lineup. New organizations need separate approval for Resonance 2; an active trial or plan does not by itself grant access. Hobby is $9/month, Builder $49/month, and Production $199/month. Spectra-2, Resonance 1, Resonance 2, Fourier, Realtime, and Proficiency share 250, 2,500, or 20,000 monthly audio minutes, respectively, including supported diarization. Each plan also includes a separate Orukeet allowance, starting at up to 20,000 transcription minutes on Hobby. Orukeet uses $0.00045/minute ($0.027/hour); optional tasks draw from its allowance at their published rates.
Assess Hume’s measurement products if
- You need offline Tagger analysis across Hume’s advertised 600+ expression dimensions and can use its contact-led access path.
- You need Prosody’s advertised real-time expression signals and have confirmed access and terms with Hume Research.
- Your application depends on Hume-specific dimensions; evaluate that need before replacing its taxonomy.
Choose oruk if
- You need to measure emotion or speaking style in recorded calls, meetings, or interviews.
- You want a compact multilabel space with published thresholding and benchmark caveats rather than hundreds of dimensions.
- You want the transcript, emotion, and style from one request, using shared subscription audio minutes.
Replacing EVI or Octave takes a different path
Oruk supplies the speech-understanding component. It does not replace EVI’s session protocol, LLM, tools, voice or audio playback. A TTS-only application needs a synthesis provider. Voice IDs and cloning rights do not transfer automatically. Use the EVI migration planning guide to inventory those dependencies; it is not a claim of a drop-in, validated full-stack replacement.
Evaluate the API you will actually use
The historical leaderboard does not establish how the current Oruk and Hume products compare. Our September API evaluation includes a downloadable input manifest, saved responses, and an offline score replay. It is a narrow, same-source diagnostic; it does not measure current Hume products or prove cross-domain accuracy.
Start with the runnable Python call-review example, then compare both services on consented recordings from your application. Record model versions, label mappings, failures, latency, and per-class errors. If you require continuous coordinates, read the distinction between valence/arousal and categorical emotion scores before treating the outputs as interchangeable.
FAQ
- Is Hume AI still an option for measuring emotion in audio files?
- Yes. Hume’s current Expression Measurement page advertises Tagger for offline batch analysis and Prosody for real-time signals. The 49.6% benchmark row comes from a dated legacy 48-dimension prosody snapshot and should not be used to infer current product availability or performance. Hume currently asks prospective Tagger users to contact its Research team for access.
- When should I choose Hume over oruk?
- Consider Hume Research if you need its advertised Tagger taxonomy or Prosody contract, and confirm access directly. Choose Oruk when one of its documented speech-analysis contracts fits your workload. EVI and TTS have a separate announced sunset; Oruk does not generate speech or reproduce an EVI session.
- How were the benchmark numbers measured?
- On speech-emotion-bench: open models were scored on the full 64,384-clip evaluation, while closed/API and audio-LLM systems were scored on a fixed 5,000-clip stratified subset. All rows use the same seven-class mapping and scorer. The oruk entry is trained in-distribution while the compared systems are evaluated zero-shot. The Hume row is a legacy 48-dimension prosody snapshot, not current Tagger or Prosody.
- Does oruk detect the same expression dimensions as Hume?
- No. Legacy Hume prosody used 48 dimensions, and Tagger advertises 600+. Oruk file models use 15 emotion and 16 style categories with model-specific score selection; Realtime has variable-length phrase arrays. Similar names do not imply equivalent scores or thresholds. Validate the exact route on your own speakers and domain.
Sources
- Hume AI’s current Expression Measurement page, reviewed October 3, 2026 — Tagger offline, Prosody real time, 600+ dimensions, 50+ languages, and contact-led access.
- Hume’s EVI and TTS sunset notice, reviewed October 3, 2026 — API access cutoff and account-data deletion notice.
- speech-emotion-bench protocol, sample counts, and downloadable per-system results: benchmark methodology. The Hume row is the legacy 48-dimension prosody endpoint, evaluated zero-shot on the 5,000-clip stratified subset; it is not a measurement of current Tagger or Prosody.
- Current oruk plan terms on the pricing page.
Migration guide from Hume EM All Hume AI alternatives oruk emotion API
Measure emotion in your own audio
Start a 7-day free trial and run the oruk API on your recordings. Plans from $9/month with audio minutes included. A card is required; $0 today. Cancel before the trial ends to avoid the subscription charge.